Presentation is loading. Please wait.

Presentation is loading. Please wait.

GFDL Hurricane Model Ensemble Performance during the 2010 Atlantic Season 65 th IHC Miami, FL 01 March 2011 Tim Marchok Morris Bender NOAA / GFDL Acknowledgments:

Similar presentations


Presentation on theme: "GFDL Hurricane Model Ensemble Performance during the 2010 Atlantic Season 65 th IHC Miami, FL 01 March 2011 Tim Marchok Morris Bender NOAA / GFDL Acknowledgments:"— Presentation transcript:

1 GFDL Hurricane Model Ensemble Performance during the 2010 Atlantic Season 65 th IHC Miami, FL 01 March 2011 Tim Marchok Morris Bender NOAA / GFDL Acknowledgments: HFIP Program, Bob Tuleya, James Franklin, Richard Pasch

2 Design of the system  Focus on intensity  Given time constraints, must be simple to implement  Development began in mid / late July  Several members created by modifying the structure- related data from the NHC storm warning message in order to perturb the axisymmetric vortex NHC 12L KATRINA 20050829 0000 272N 0891W 335 046 0904 1006 0649 72 037 0371 0334 0278 0334 D 0204 0185 0139 0185 72 410N 815W 0167 0167 0093 0167 Radii of 34- and 50-kt winds ROCI

3 Defining the ensemble members  11-member ensemble: 10 perturbed members and a control forecast  GP0: Control forecast (GFD5 run on Jet)  GPA: Unbogussed forecast  GPB: Control, but with no asymmetries included  GPC: Control, but with the use of old environmental filter  GPD: Increase storm size (ROCI-based) by 25%  GPE: Decrease storm size (ROCI-based) by 25%  GPF: Increase wind radii 25%, increase storm size 25%  GPG: Decrease wind radii 25%, decrease storm size 25%  GPH: Old filter (GPC), plus both size increases from GPF  GPJ: Old filter (GPC), plus both size decreases from GPG  GPK: Set Rmax minimum to 45 km (GFD5 control uses 25 km) Size Members

4 Running the ensemble  All 11 members run on FSL / HFIP Jet system at operational resolution (1/2 o, 1/6 o, 1/12 o )  Each member’s 126-h forecast ran in 90 minutes using 31 cpus.  Full forecast cycle in under 2 hours, allowing for up to 3 storms 4x per day.  Automatic cycling of the system began on 21 September

5 Cases  195 cases  Alex (12 cases)  Danielle (16)  Earl (20)  Igor (24)  Julia (15)  Lisa (15)  Matthew (11)  Nicole (2)  Otto (9)  Paula (15)  Richard (19)  Shary (4)  Tomas (33)

6 Earl 2010083100 Modifications to the observed storm size data that are input into the axisymmetric spinup lead to noticeable changes in the model storm structure GPF: Increase all wind radii and storm size (ROCI) by 25% GPG: Decrease all wind radii and storm size (ROCI) by 25%

7 Earl 2010083100 Spread evident in the V T profiles from the vortex spinup leads to noticeable spread in the Vmax forecast, but very little spread in the tracks.

8 Igor 2010090912 As a weak tropical storm (35 kts), there is little in the way of structure to manipulate for this case of Igor.

9 Igor 2010090912 Small spread is seen throughout this case for both track and intensity

10 Igor Later cases of Igor – when more intense – exhibited more spread, at least for the intensity forecasts. 2010091300 (130 kts)2010091312 (130 kts)

11 Paula Cases of Paula exhibited some of the most spread for track forecasts

12 Nicole A problem that occasionally occurred with very weak storms is shown here for Nicole in which several identical forecasts were produced.

13 Intensity Results Statistically significant improvements of the ensemble mean over the control are seen through the middle of the forecast period. 4.4% 8.4% 8.3% 11.2% 11.8% However, the spread results indicate an underdispersive ensemble

14 Intensity Results Improvements over the operational GFDL are also seen, especially at 4 and 5 days. 4.4% 8.4% 8.3% 11.2% 11.8%

15 Intensity Results Verification rank histograms also indicate an underdispersive ensemble.

16 Intensity Improvement Stratified by Vmax (0) Forecasts for weak storms see very little improvement from the GFDL ensemble mean until 3 days, while stronger storms see improvements at earlier lead times.

17 Track Results Improvements for track are smaller than for intensity, but still significant from 12-48h. However, the spread in the track forecasts is extremely low. 5.5% 6.3% 3.6% 4.1% 1.0%

18 Performance of the unbogussed forecast An interesting result was how well the unbogussed forecast performed, both for track and intensity. Track: Unbogussed forecast had smallest error at 24, 36, 48 and 96h. Frequency of Superior Performance: Track

19 Comparison of Intensity Forecast Biases Modifying the wind radii and storm size together has more of an impact on intensity bias than just modifying storm size alone.

20 Testing a modified ensemble membership for 2011  13-member ensemble: 12 perturbed members and a control forecast  0: Control forecast (Proposed 2011 operational GFDL)  A: Unbogussed forecast  B: Increase Vmax (0) +10%  C: Decrease Vmax (0) -10%  D: Increase 34R (+25%), 50R (+40%), ROCI (+25%)  E: Decrease 34R (-25%), 50R (-40%), ROCI (-25%)  F: Increase Rmax +25%  G: Decrease Rmax -25%  H: GFDL ensemble control from 2010 (“GFD5”)  J: Allow greater % of dissipation to go into heating  K: Decrease the amount of vertical momentum transport  L: Turn off shallow convection  M: Reduce the penetration of downdrafts into the boundary layer Size Members Convective Param Members

21 Differences with convective perts: Tomas example Considerable spread in the intensity forecast is seen from the 4 convective parameterization members for this case of Tomas.

22 Preliminary Results: Track 2011 ensemble mean better than 2011 control and significantly better than 2010 ensemble mean at 3-5 days lead time. Ensemble only marginally more dispersive than 2010 ensemble.

23 Preliminary Results: Intensity 2011 ensemble mean better than control but significantly worse than 2010 ensemble mean at 3-5 days lead time.

24 Conclusions  Modest improvements over control run seen for 2010 ensemble mean intensity forecasts, somewhat less for track.  Modifying the vortex structure shows promise for perturbing forecasts (although not as well for weak storms, given the perturbation method we used).  The ensemble shows low dispersion for intensity and especially track, and it may benefit from increased spread.  Modifying the wind radii and the ROCI has more of an impact on intensity bias than just modifying ROCI alone.  The unbogussed forecast performed surprisingly well for track, and even for intensity at later lead times.


Download ppt "GFDL Hurricane Model Ensemble Performance during the 2010 Atlantic Season 65 th IHC Miami, FL 01 March 2011 Tim Marchok Morris Bender NOAA / GFDL Acknowledgments:"

Similar presentations


Ads by Google